Voynich | edkSG Vol. 1 — The Research Baseline

What we know, what remains unproved, and the pathway from here

Edition date: 5 September 2026 · Singapore
Volume ID: EDKSG-VOY-V001 · Edition: 1.0.0
Baseline boundary: work undertaken before 5 September 2026, 00:00, Asia/Singapore
Collection owner: eduKateSG Voynich Research Library
Editorial framework: Wintour House | The eduKate Publishing House

A research project needs somewhere to stand. Not somewhere from which every question has been answered, but somewhere from which the next question can be asked without rebuilding the entire investigation.

This is that place for our Voynich work.

The earlier research becomes the inherited baseline. Its observations remain observations. Its hypotheses remain hypotheses. Its failed interpretations remain in the record. Its unfinished tests do not become finished because we have opened a new volume.

From here, the programme has a dated pathway. Each new volume must explain what it inherited, what question it investigated, what evidence it examined, what changed, and what remains unresolved. The reader should be able to follow the investigation forward. A later researcher should be able to trace a conclusion backwards without depending on our memory of how we reached it.

The baseline is closed as a historical starting point. The manuscript is not declared solved.

That distinction is the foundation of this edition.

Contents — follow the research pathway

1. What “the baseline is done” means
2. One existing library, a new chronological series
3. The object comes before our explanation of it
4. The strongest inherited structural result
5. What the structural result does not establish
6. The physical sequence remains a research question
7. The writing has provisional structural leads
8. Before translating words, establish the units
9. The visual experiment is still an open gate
10. Padua and the comparators retain their proper role
11. What the Tangential Lens is allowed to do
12. The inherited claim register
13. A catalogue must record change, not manufacture progress
14. How the edition numbering will work
15. What the next research record should contain
16. The future connection between text and image
17. What would count as a decipherment
18. How we will recognise a useful negative result
19. What this first volume changes
20. The pathway begins here

1. What “the baseline is done” means

For this series, the boundary is exact: 5 September 2026 at 00:00 Singapore time. Work undertaken before that boundary belongs to the legacy baseline, designated EDKSG-VOY-B000. Work undertaken on or after it belongs to the continuing record.

The date is an administrative boundary, not a scientific verdict. It prevents the project from continuously moving its starting line. We no longer need to pretend that every new research session begins with an untouched manuscript and no prior work. We have a history. The purpose of the baseline is to preserve it honestly.

There are two different senses of completion here. The first is closure of a period: the work from that period can be cited as the inherited state. The second is completion of an investigation: a particular test has run, its outputs have been examined, and its result has been recorded. Closing the period does not complete every investigation within it.

An unfinished visual experiment therefore crosses the boundary as an unfinished experiment. A proposed meaning for a glyph crosses it as a proposed meaning. An inaccessible source crosses it as an access gap. None receives an evidential promotion merely because it now occupies a numbered shelf.

Nor does the boundary certify that every earlier conversation, unpublished note, scan or article has been recovered into this package. The public catalogue and the principal technical records provide the starting references. Their coverage limits travel with them. An earlier item recovered later can be added through a dated legacy accession. Its original date and its accession date must both be retained.

This gives the project continuity without pretending to have perfect archival completeness.

2. One existing library, a new chronological series

The Voynich Research Library already provides the public subject catalogue. Its section labelled as verified on 4 September 2026 reports 303 published articles, distributed across five collections. That is the catalogue’s reported audit state, not a new independent count of all live WordPress records performed for this volume. [1]

Those entries keep their existing titles, URLs and catalogue positions. This volume does not replace the hub, reassign its entry 001, or turn the earlier articles into chapters that must now be rewritten. The hub remains the subject doorway. The new volumes provide a chronological account of changes in the research state.

The distinction is useful. A reader looking for the treatment of foldouts should still use the subject collection. A reader asking what changed between two research checkpoints should use the dated volumes. The same source can be relevant to both routes without becoming two independent pieces of evidence.

The identifiers therefore do different jobs. EDKSG-VOY-V001 identifies this editorial work. Its edition number identifies a particular version. An inherited test identifier still identifies its test. A source identifier still identifies its source. The volume prefix is a local editorial convention, not a claim that a new production identity has been allocated in another eduKate system.

The new series adds a timeline to the library. It does not create a rival library.

3. The object comes before our explanation of it

The manuscript at the centre of the investigation is Beinecke MS 408, held by Yale’s Beinecke Rare Book and Manuscript Library. Yale describes an unidentified script and an undeciphered text, and provides high-resolution digital images for research. Its account also distinguishes the surviving object’s material evidence from the many interpretations attached to it. [2]

That external anchor matters because our research library is not the manuscript. It is a collection of representations, descriptions, comparisons and arguments about the manuscript.

A photograph gives access to what that photograph records. A transcription gives access to decisions about the marks visible in a representation. A statistical model gives access to patterns under its input definitions. A fluent explanation gives access to an interpretation of those patterns. None of these should silently replace the object from which the investigation began.

For this series, the practical rule is simple: a claim about physical construction requires evidence appropriate to physical construction. A claim about transcription requires the relevant transcription and its uncertainty rules. A claim about meaning requires more than the fact that a pattern exists.

This is not a reason to distrust every representation. It is a reason to use each representation for the questions it can answer. A reliable measurement of line position can be valuable even when it cannot identify a language. A high-quality image can support a visual comparison without resolving a question about the three-dimensional binding.

The most expensive mistake would be to perform increasingly sophisticated analysis while forgetting which layer supplied the data.

4. The strongest inherited structural result

The main technical starting point is the Voynich Test 1b Frozen Segmentation Matrix, frozen on 2 August 2026. It describes the manuscript through separate axes rather than assigning one explanation to every kind of difference. The archived workbook records 227 analytical page units, 52 represented bifolios, statistical-cluster coverage for 225 page units, and 47 recorded boundary events including START. [3]

These numbers describe the analytical dataset. The 227 units must not be substituted for the physical manuscript’s page count. Foldout panels, transcription coverage and page-unit conventions belong to the definition of the dataset.

The axes include physical grouping, scribal assignment, Currier classification, broad visual form, placement of writing, and a statistical grouping of transcribed forms. Their purpose is to keep different kinds of organisation visible at the same time.

The workbook reports the following comparisons within bifolios:

These are inherited workbook results, not experiments rerun for this edition. The denominators differ because the available assignments differ. A comparison based on 45 completely assigned bifolios is not a claim that all 52 represented bifolios were completely classified. [3]

The central finding is descriptive: the recorded layers do not all coincide. A strong association between two layers does not make them identical, and a change in one layer does not require every other layer to change with it.

For the continuing investigation, this means that a proposed explanation must state which organisation it is explaining. A model of handwriting distribution should not automatically be presented as a model of subject matter. A visual grouping should not automatically be presented as a set of translated chapters.

The baseline gives us a structured comparison problem. It does not give us a decoded table of contents.

5. What the structural result does not establish

The workbook’s own source notes make an important qualification: its broad visual categories recode existing source assignments, while its hand and Currier fields inherit prior scholarly classifications. Some of these variables are also dependent. Their agreement is therefore not equivalent to several wholly independent discoveries. [3]

Neutral terminology is still useful. Describing a form before naming its function can reduce interpretive pressure. But replacing a conventional name with a neutral code does not, by itself, constitute a blind recovery of the manuscript’s visual structure.

The same caution applies to statistical clusters. The workbook explicitly treats them as distributional classes, not decoded topics. It also records an assumption that the same transcribed form can be treated consistently across pages. That assumption may be useful for analysis, but its adequacy remains part of what needs testing. [3]

Consider a deliberately hypothetical example. Suppose pages with one kind of illustration contain more of a certain token family. That association could reflect subject matter. It could also reflect a scribe, a copying source, a page-layout convention, a text-generation procedure, or another variable that travels with the pictures. The association alone does not choose among those explanations.

This is why we retain the matrix rather than appointing one of its columns as the hidden truth behind all the others. The rejected claim is that the tested simple partitions already establish one unique functional organisation. We are not asserting that no possible mathematical partition could encode the observed combinations.

A description can become more precise while the function remains unknown. That is the state we inherit.

6. The physical sequence remains a research question

Lisa Fagin Davis’s codicological work examines the relationship between scribal assignments and the manuscript’s folded sheets, numbering, stains and binding history. She discusses evidence favouring rearrangement through rebinding, while preserving uncertainty about the original sequence. This is external manuscript scholarship incorporated into our investigation, not an eduKate discovery. [4]

Its methodological consequence is substantial. Present adjacency should not be treated as an automatic record of original explanatory order.

A model may find that one page predicts the next page under today’s sequence. That can be an interesting observation. Before interpreting it as the author’s intended progression, however, the model must account for the possibility that the sequence changed. Conversely, failure to find a smooth present-day progression does not show that the original work lacked one.

Future sequence tests should therefore distinguish the observed order from proposed reconstructions. They should preserve the physical constraints used to justify each reconstruction rather than arranging pages until a preferred narrative becomes persuasive.

A useful test would report what remains stable across plausible orderings and what depends entirely on one ordering. This volume does not report such a test as completed. It records why that test is needed and why the existing order cannot be made an invisible assumption.

The book’s history is part of the problem, not background scenery around the text.

7. The writing has provisional structural leads

The archived Voynich Decoder Living Runtime, version 0.1.0, dated 3 August 2026, separates the text investigation from the visual investigation. Its text record reports locally conditioned choices between forms represented as oR and qoR, contributions from preceding terminal forms, and strong line-boundary effects. It marks these observations as provisional and nonsemantic. [5]

Here, the symbols are analytical notation. They do not establish a decoded root, a spoken form, or a lexical meaning. In particular, the runtime does not authorise q as “use”, “apply”, “enter”, “route”, “select”, or a universal operator. [5]

That restraint is valuable. An apparently small change at a boundary could arise from several kinds of process. It might be connected with writing conventions, text construction, copying, abbreviation, encoding, or language. A positional regularity does not choose its own explanation.

The next text question in that record concerns matched forms with and without final y. It asks whether the ending behaves as a reproducible structural indicator under controlled comparison. The record requires controls for the underlying form, placement, page class, scribal or transcription regime, and prediction on material not used to construct the proposal. [5]

This volume inherits the question. It does not announce the answer.

Nor does it treat older prose saying that an effect was “replicated” as sufficient evidence of independent replication. Two representations of the same underlying manuscript can help test sensitivity to transcription choices. They do not automatically supply independent historical evidence or independent analytical judgement.

The productive next step is to recover the exact inputs and outputs required to test the claim. Another attractive gloss for q would not close that gap.

8. Before translating words, establish the units

A central question for the continuing programme is whether the units we count are adequate representations of the units used by the manuscript’s writing process.

This is a research question, not a new finding announced in this volume. A visible gap may be an appropriate analytical separator without already being established as a linguistic word boundary. A repeated glyph shape may be a useful transcription unit without already being established as a letter representing a sound.

The distinction matters whenever we compare models. A model can perform poorly because its proposed language is wrong. It can also perform poorly because its input segmentation is wrong. Unless those possibilities are separated, failure at one level may be incorrectly assigned to another.

The first obligation is therefore to preserve ambiguity rather than normalise it away unnoticed. Alternative transcriptions, uncertain joins, differing gap classes and layout context should remain recoverable from the analysis. A result that depends on one disputed transcription choice should say so.

The second obligation is to compare candidates under shared conditions. A candidate that is allowed to move boundaries after inspecting the desired reading has more freedom than one required to use fixed boundaries. Their apparent successes are not directly comparable.

The third obligation is to test generalisation. A proposed representation should do useful work on material that was not used to invent or tune it. Otherwise, it may merely describe the choices that made it look successful.

For our pathway, unit recovery is therefore a priority. It is not a declaration that visible spaces are meaningless, that Voynich has no words, or that a particular cipher has been identified.

9. The visual experiment is still an open gate

The retrieved decoder record labels Test 2A as ACTIVE_NOT_COMPLETED. Its stated purpose is to determine whether recurring visual components can be identified blindly, reproducibly, and above matched-control expectations. The record specifies anonymised crops and neutral component descriptions rather than premature semantic names. [5]

That is the inherited status. No completed superseding result was established for this edition.

The question is narrower than “What are the plants?” It asks whether a recurring component vocabulary can be recovered before deciding what the components represent. An axial form, branching event, collar-like band, attachment geometry or terminal shape can be coded structurally without immediately identifying a species, organ, ingredient or machine part.

Even a positive result would have a limited meaning. It would show that the proposed components recur under the stated method and controls. It would not automatically show that their recurrence encodes instructions, medicine, commerce, or a particular historical tradition.

A negative result would also need careful wording. Failure to recover one proposed component scheme would count against that scheme under the test conditions. It would not establish that the manuscript has no visual organisation at all.

The completion record must therefore include both matches and failures, the comparison material, the treatment of ambiguous crops, and the coding procedure. Repeated judgement by the same analyst may test consistency, but should not be advertised as independent expert agreement.

The next volume may prepare or complete part of this work. Its title must say which occurred. “Protocol”, “corpus audit”, “pilot”, “result” and “replication” are not interchangeable labels.

10. Padua and the comparators retain their proper role

The inherited Padua control record explicitly separates historical comparison from Voynich identity. It does not establish Voynich’s provenance, ownership, language, function, or direct membership in a Paduan production environment. Its permitted use is bounded comparison. [6]

This leaves the comparator work with an important job. It helps ask what a known kind of manuscript actually does, what evidence normally establishes that function, and which visible similarities fail to discriminate between competing explanations.

A known manuscript may reveal how labels attach to diagrams or how entries are arranged. Such a comparison can generate a prediction about Voynich. It cannot confer its own identity on Voynich merely because some visual features resemble each other.

The missing step is discriminating evidence. A feature that occurs in many environments does little to select one environment. A proposed localisation must confront alternatives, not simply accumulate details compatible with the preferred place.

The same applies to period-aware looking. For the ongoing pathway, a broad comparison window around 1400 can be useful as a search design. It is not a newly established date of composition, and it must not be confused with the dating of particular materials. A period resemblance is a candidate clue whose value depends on its specificity and transmission history.

We retain Padua, Carrara, Masson and other comparison routes. We do not retain the right to use resemblance as a shortcut around missing evidence.

11. What the Tangential Lens is allowed to do

The Tangential branch changes the observer’s assumptions by placing the unknown object inside deliberately incompatible explanatory settings. Its value is methodological: it can expose questions that familiar labels prevented us from asking. The existing library presents that branch as a generator of tests, not a set of historical identifications. [1]

Imagine, as a thought experiment, treating a repeated form as a routing mark rather than a noun. That can suggest a test for permitted and forbidden transitions. Treating page organisation as a record system can suggest a test for repeated fields. Neither exercise establishes that the manuscript is literally a control system or a database.

The useful output is the testable residue after the borrowed world has been removed.

A proposed transition rule must therefore be expressible without relying on the metaphor that inspired it. The rule should also face a competing explanation that might generate the same transition pattern. Otherwise, the metaphor has merely renamed the mystery.

There is a second danger: different imagined worlds can share very general concepts such as repetition, hierarchy and connection. Agreement between those descriptions may reflect the generality of the concepts rather than independent support for a manuscript theory.

The continuing volumes will distinguish a new lens from a new observation. A lens can earn a place in the archive because it produced a useful question. It earns an evidential upgrade only when the resulting test has actually been performed and its outcome survives the relevant controls.

Changing the observer is permitted. Changing the evidence to fit the observer is not.

12. The inherited claim register

The following register preserves the identities of the principal claims used in the technical baseline. These are not newly minted discoveries. They are a compact crosswalk into the inherited record. Capitalisation variants in earlier records should be treated as aliases, not additional claims. [3, 5]

This table is deliberately less dramatic than a list of discoveries. It shows where the programme can safely place its weight.

There is also a wording correction worth making explicit in the new series. Some inherited materials use the slogan that interpretations remain false until they survive an attempt to break them. The useful intention is adversarial testing. The precise evidential status before such a test is usually unestablished, not necessarily false. Surviving one test is also not proof of truth.

The new volumes will use those distinctions. The older wording is preserved as part of the historical record rather than silently rewritten. This is an editorial clarification of our method, not a new finding about the manuscript.

13. A catalogue must record change, not manufacture progress

A dated article can record research, but its existence cannot be the evidence that the research succeeded.

Each volume will therefore state its entry condition: the specific prior record from which it begins. It will identify the question being examined, the evidence and method used, the observed outcome, and the next justified action. When no experiment has been run, the outcome field must say that the work was preparatory or interpretive.

Several legitimate volume types follow from this rule. A corpus audit can establish which inputs are available. A protocol can specify a test. A pilot can expose problems in the measurement procedure. A result can report an executed comparison. A replication can examine whether a prior outcome survives repetition under stated conditions. A correction can weaken an earlier claim.

None is a lesser contribution merely because it does not announce a solution. Their value depends on the uncertainty they remove and the integrity of the handoff they create.

The record should also distinguish three dates: when the underlying work occurred, when its evidence was observed or retrieved, and when the account was issued. This is particularly important at the baseline boundary. A source retrieved on 5 September may document work from August. Its retrieval belongs to the new record; the older work retains its earlier date.

A later correction does not retroactively turn the old volume into an account of knowledge that did not yet exist. It becomes a linked successor. The original remains readable as the state of the investigation at that time.

14. How the edition numbering will work

The public series title is Voynich | edkSG Research Volumes. The first work is Voynich | edkSG Vol. 1 — The Research Baseline, with the stable volume identifier EDKSG-VOY-V001.

The next volume will receive the next number only when its reader job and content are defined. This edition reserves the direction of travel, not a fictional publication. There is no live Volume 2 claimed in this package.

A volume number identifies a work in the sequence. An edition number identifies a revision of that work. A publication date records an actual release. These should not be collapsed into one label. A prepared file can have an edition date while its publication date remains empty.

Minor corrections can be issued as a revised edition with an explicit change note. A new test or a materially new research question normally deserves a new volume, linked to its predecessor. A claim that changes status should retain its identity and gain a dated status event. Renaming the claim to hide its earlier failure would destroy the pathway we are building.

Legacy articles need not be republished to join this structure. They can remain where they are, linked through their existing catalogue positions and source references. The dated series explains how those sources bear on the current research state.

This arrangement gives us stable references without freezing thought. It makes revision possible without making history disappear.

15. What the next research record should contain

The immediate priority is an input and protocol audit for unit and boundary testing, coordinated with the already open visual-component gate. This is the proposed next research job, not a completed outcome.

For the text lane, the record should name the exact transcription versions, the treatment of uncertain marks, and the definitions of separators and analytical units. It should recover the inputs and calculations behind the inherited boundary claims before describing those claims as independently reproduced.

For the visual lane, it should identify the source images and crop rules needed by Test 2A. A crop set needs a documented relationship to its source images, but the coding stage should conceal identifying information where the protocol requires blindness. The later assessment must be able to recover the original locations to check whether a supposed match survived context.

The test design should specify how physically or analytically related material is kept from leaking across development and evaluation. A method tuned on one side of a tightly related group should not gain a misleading advantage by being evaluated on near-duplicate material from that same group.

Thresholds, comparison classes and exclusions should be fixed before the decisive outcomes are inspected. Exploratory changes are allowed, but they create a revised exploratory analysis rather than a retroactively preregistered success.

A useful first output may simply be an honest availability report: these inputs are present, these are incomplete, these are ambiguous, and these are required before the test can run. That would advance the pathway more than a confident interpretation resting on unlocated calculations.

16. The future connection between text and image

Once the two lanes have independently established usable patterns, the programme can test whether a visual component predicts a textual feature beyond what is explained by other variables.

The word independently matters. A text family selected because it already seemed to match a picture should not then be presented as independent confirmation of that picture’s meaning. The selection history belongs in the record.

A proposed cross-lane test should compare a clearly specified visual feature with a clearly specified textual feature while accounting for plausible confounders such as scribal assignment, broad page class, text placement and position. It should then evaluate the relationship on material not used to discover it.

There are at least three informative outcomes. A relationship may remain after the controls, disappear after them, or remain too uncertain to distinguish from alternatives. Each outcome creates a different next step. A surviving relationship justifies closer investigation; a disappearing relationship weakens the original interpretation; an inconclusive relationship identifies a measurement or sample problem.

None of these outcomes automatically names a plant, identifies a language, or translates a sentence.

The achievement would be more fundamental: a reproducible relationship between two independently described parts of the manuscript. Meaning could then be investigated from a firmer starting point instead of being inserted at the beginning.

This is the bridge the pathway should try to earn.

17. What would count as a decipherment

This series will not use “solved” as a reward for coherence. A proposed linguistic decipherment would need a stable account of its units, defensible reading rules, justified meanings, substantial coverage, and successful application beyond the examples used to construct it.

A few plausible words are not enough when the procedure permits flexible substitutions, rearrangements or selective omissions. Such freedom must be declared and constrained. Otherwise, the apparent translation may be the product of the interpreter’s choices rather than the manuscript.

The proposed reading should also account for inconvenient material. Differences between textual populations, unusual placements, short labels and longer passages cannot simply be dismissed whenever they fail to fit. Their treatment belongs to the explanation.

Independent reproduction matters because the method should work when another competent researcher applies the stated rules without private coaching toward a desired answer. Disagreement should be recorded at the level of inputs and rules, not hidden behind a single success label.

A non-linguistic explanation would face a related but different burden. Generating something similar is evidence that a mechanism is possible. Establishing that it best explains this manuscript requires comparison against alternatives and attention to the actual object’s features and history.

These are proposed success conditions for the continuing programme. They do not imply that the solution must take a familiar linguistic form. They make clear what kind of evidence would justify changing our present status.

18. How we will recognise a useful negative result

A pathway is not useful if every turn is described as a breakthrough. Some of the most valuable volumes may show that a promising route does not carry the weight assigned to it.

Suppose a proposed boundary effect disappears when line position is matched. That would weaken the claim that the preceding token supplies additional predictive information. It would not make the work worthless: the result would identify the variable that had been doing the explanatory work.

Suppose recurring visual forms are no more reliably matched than the negative controls. That would count against the tested component scheme, not prove that every visual resemblance is meaningless. The record would preserve the crop method and the failed scheme so that a later investigator does not repeat it unknowingly.

Suppose a historical comparison remains compatible with several regions. The correct result would be unresolved localisation, not selection of the region with the most appealing narrative. That outcome protects the next search from inheriting a false certainty.

These examples are hypothetical. No such new outcomes are reported in this volume. They illustrate the language that later records should use.

The point is to make failure specific enough to be reusable. “It did not work” is too vague. “This prediction failed under these inputs and controls” is an asset to the investigation.

19. What this first volume changes

This volume changes the organisation of the research record, not the decipherment status of the manuscript.

It establishes an explicit temporal baseline. It separates the existing subject catalogue from the new chronological series. It preserves the central inherited claim identities. It records open tests as open. It distinguishes reported results from results rerun in this edition. It identifies a next job whose completion can be checked.

It also limits what this edition claims about its own production. The technical summaries here are based on retrieved archival records. The earlier tests have not all been independently repeated. The public catalogue’s reported count is not presented as a fresh complete site audit. The source list below is the consulted foundation for this volume, not an assertion that every prior research source has been reacquired.

These limits are part of the record’s usefulness. They tell the next researcher where to verify, not merely what to believe.

The editorial work follows the published Wintour House distinction between a manuscript, an edition and a published manifestation. The framework requires evidence boundaries, identifiable editions, correction behaviour and a return to the originating collection. Following that framework is not a claim that named individuals or autonomous editorial agents participated in this preparation. [7]

20. The pathway begins here

A long investigation can lose its way in two opposite directions. It can discard too much and repeatedly start again. Or it can preserve so much unexamined confidence that every later question is forced to agree with an earlier story.

The baseline is intended to avoid both mistakes.

We retain the work. We retain its limits. We retain the possibility that some of our most attractive interpretations will fail. We do not ask future research to defend the prestige of the archive. We ask it to improve the archive’s contact with the manuscript.

The next volume should leave a visible difference: a recovered input, a clarified unit, a tested prediction, a rejected interpretation, a reproduced result, or a better-defined unknown. It should not merely add another layer of prose around the same uncertainty.

Before 5 September 2026: the inherited baseline. From 5 September 2026 onward: the dated pathway.

The base is established as a record. The questions remain alive.

---

Source notes and reading routes

[1] Existing public catalogue. Voynich Research Library, particularly “The Complete Voynich Article Library”, labelled verified 4 September 2026. Consulted 5 September 2026. The count and collection totals above are attributed to that catalogue; individual linked articles and WordPress records were not all re-audited for this edition.

[2] Institutional object record. Yale Beinecke Library: The Beinecke Cipher (Voynich) Manuscript. Consulted 5 September 2026. Used for the manuscript identifier, institution, unread status and digital-access description, not as a blanket resolution of contested chronology or provenance details.

[3] Archived internal research result. Voynich_Test_1b_Frozen_Segmentation_Matrix.xlsx, frozen 2 August 2026. Relevant sheets: README; Frozen Matrix; Cross-Axis; Defeat Ledger; Codebook & Sources. Retained in the eduKate research archive. Numerical summaries are inherited reports. The source classifications, their dependencies, coverage and token-identity assumptions remain attached to the results.

[4] External codicological research. Lisa Fagin Davis, “Voynich Codicology”, 19 January 2025. Consulted 5 September 2026. Used as an attributed scholarly interpretation of physical and scribal evidence, not as a newly performed object examination.

[5] Archived internal research state. voynich-decoder-living-runtime-v0.1.html, version 0.1.0, dated 3 August 2026. Relevant sections: frozen Test 1b; claim registry; provisional text lane; Test 2A; limitations and continuation conditions. Its “decoder” title names the research framework; its status fields do not claim plaintext translation.

[6] Archived comparative control. padua-parent-city-bottom-runtime-v1.html, public-history boundary and permitted comparison conditions. Used only for the project’s explicit restrictions on Voynich identity and provenance claims. The historical material itself is not reproduced here.

[7] Editorial framework. Wintour House | The eduKate Publishing House, especially the edition relationships, evidence boundaries, publication states and attribution boundary. Consulted 5 September 2026.

Existing reading routes. The Voynich Research Library remains the canonical doorway. Its existing Voynich master article and Mathematics and Science synthesis remain separate works; this volume does not replace them.

Edition record

EDKSG-VOY-V001 · Edition 1.0.0 · Published 5 September 2026 · Baseline EDKSG-VOY-B000.

No new decipherment, completed Test 2A, completed final-y test, independent replication, complete archival census, or production-system ingestion is claimed by this edition. Subsequent corrections or results must identify their relationship to this record and preserve its historical state.